Re: [PATCH] vdpa/mlx5: Fix crash on shutdown for when no ndev exists

From: Dragos Tatulea
Date: Wed Aug 02 2023 - 03:59:39 EST

Next message: David Rheinsberg: "Re: [PATCH] memfd: support MFD_NOEXEC alongside MFD_EXEC"
Previous message: Daniel Baluta: "Re: [PATCH 1/2] ASoC: simple-card: Introduce playback-only/capture only DAI link flags"
In reply to: Jason Wang: "Re: [PATCH] vdpa/mlx5: Fix crash on shutdown for when no ndev exists"
Next in thread: Dragos Tatulea: "Re: [PATCH] vdpa/mlx5: Fix crash on shutdown for when no ndev exists"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On Wed, 2023-08-02 at 10:51 +0800, Jason Wang wrote:
> On Tue, Aug 1, 2023 at 4:17 PM Dragos Tatulea <dtatulea@xxxxxxxxxx> wrote:
> >
> > On Tue, 2023-08-01 at 11:59 +0800, Jason Wang wrote:
> > > On Mon, Jul 31, 2023 at 5:08 PM Michael S. Tsirkin <mst@xxxxxxxxxx> wrote:
> > > >
> > > > On Mon, Jul 31, 2023 at 07:15:31AM +0000, Dragos Tatulea wrote:
> > > > > On Thu, 2023-07-27 at 12:28 -0400, Michael S. Tsirkin wrote:
> > > > > > On Thu, Jul 27, 2023 at 04:02:16PM +0000, Dragos Tatulea wrote:
> > > > > > > On Wed, 2023-07-26 at 15:26 -0400, Michael S. Tsirkin wrote:
> > > > > > > > On Wed, Jul 26, 2023 at 10:07:38PM +0300, Dragos Tatulea wrote:
> > > > > > > > > The ndev was accessed on shutdown without a check if it
> > > > > > > > > actually
> > > > > > > > > exists.
> > > > > > > > > This triggered the crash pasted below. This patch simply adds
> > > > > > > > > a
> > > > > > > > > check
> > > > > > > > > before using ndev.
> > > > > > > > >
> > > > > > > > > BUG: kernel NULL pointer dereference, address:
> > > > > > > > > 0000000000000300
> > > > > > > > > #PF: supervisor read access in kernel mode
> > > > > > > > > #PF: error_code(0x0000) - not-present page
> > > > > > > > > PGD 0 P4D 0
> > > > > > > > > Oops: 0000 [#1] SMP
> > > > > > > > > CPU: 0 PID: 1 Comm: systemd-shutdow Not tainted 6.5.0-
> > > > > > > > > rc2_for_upstream_min_debug_2023_07_17_15_05 #1
> > > > > > > > > Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-
> > > > > > > > > 1.13.0-0-
> > > > > > > > > gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
> > > > > > > > > RIP: 0010:mlx5v_shutdown+0xe/0x50 [mlx5_vdpa]
> > > > > > > > > RSP: 0018:ffff8881003bfdc0 EFLAGS: 00010286
> > > > > > > > > RAX: ffff888103befba0 RBX: ffff888109d28008 RCX:
> > > > > > > > > 0000000000000017
> > > > > > > > > RDX: 0000000000000001 RSI: 0000000000000212 RDI:
> > > > > > > > > ffff888109d28000
> > > > > > > > > RBP: 0000000000000000 R08: 0000000d3a3a3882 R09:
> > > > > > > > > 0000000000000001
> > > > > > > > > R10: 0000000000000000 R11: 0000000000000000 R12:
> > > > > > > > > ffff888109d28000
> > > > > > > > > R13: ffff888109d28080 R14: 00000000fee1dead R15:
> > > > > > > > > 0000000000000000
> > > > > > > > > FS: 00007f4969e0be40(0000) GS:ffff88852c800000(0000)
> > > > > > > > > knlGS:0000000000000000
> > > > > > > > > CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> > > > > > > > > CR2: 0000000000000300 CR3: 00000001051cd006 CR4:
> > > > > > > > > 0000000000370eb0
> > > > > > > > > DR0: 0000000000000000 DR1: 0000000000000000 DR2:
> > > > > > > > > 0000000000000000
> > > > > > > > > DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7:
> > > > > > > > > 0000000000000400
> > > > > > > > > Call Trace:
> > > > > > > > > <TASK>
> > > > > > > > > ? __die+0x20/0x60
> > > > > > > > > ? page_fault_oops+0x14c/0x3c0
> > > > > > > > > ? exc_page_fault+0x75/0x140
> > > > > > > > > ? asm_exc_page_fault+0x22/0x30
> > > > > > > > > ? mlx5v_shutdown+0xe/0x50 [mlx5_vdpa]
> > > > > > > > > device_shutdown+0x13e/0x1e0
> > > > > > > > > kernel_restart+0x36/0x90
> > > > > > > > > __do_sys_reboot+0x141/0x210
> > > > > > > > > ? vfs_writev+0xcd/0x140
> > > > > > > > > ? handle_mm_fault+0x161/0x260
> > > > > > > > > ? do_writev+0x6b/0x110
> > > > > > > > > do_syscall_64+0x3d/0x90
> > > > > > > > > entry_SYSCALL_64_after_hwframe+0x46/0xb0
> > > > > > > > > RIP: 0033:0x7f496990fb56
> > > > > > > > > RSP: 002b:00007fffc7bdde88 EFLAGS: 00000206 ORIG_RAX:
> > > > > > > > > 00000000000000a9
> > > > > > > > > RAX: ffffffffffffffda RBX: 0000000000000000 RCX:
> > > > > > > > > 00007f496990fb56
> > > > > > > > > RDX: 0000000001234567 RSI: 0000000028121969 RDI:
> > > > > > > > > fffffffffee1dead
> > > > > > > > > RBP: 00007fffc7bde1d0 R08: 0000000000000000 R09:
> > > > > > > > > 0000000000000000
> > > > > > > > > R10: 0000000000000000 R11: 0000000000000206 R12:
> > > > > > > > > 0000000000000000
> > > > > > > > > R13: 00007fffc7bddf10 R14: 0000000000000000 R15:
> > > > > > > > > 00007fffc7bde2b8
> > > > > > > > > </TASK>
> > > > > > > > > CR2: 0000000000000300
> > > > > > > > > ---[ end trace 0000000000000000 ]---
> > > > > > > > >
> > > > > > > > > Fixes: bc9a2b3e686e ("vdpa/mlx5: Support interrupt bypassing")
> > > > > > > > > Signed-off-by: Dragos Tatulea <dtatulea@xxxxxxxxxx>
> > > > > > > > > ---
> > > > > > > > > drivers/vdpa/mlx5/net/mlx5_vnet.c | 3 ++-
> > > > > > > > > 1 file changed, 2 insertions(+), 1 deletion(-)
> > > > > > > > >
> > > > > > > > > diff --git a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > > > > > > > > b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > > > > > > > > index 9138ef2fb2c8..e2e7ebd71798 100644
> > > > > > > > > --- a/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > > > > > > > > +++ b/drivers/vdpa/mlx5/net/mlx5_vnet.c
> > > > > > > > > @@ -3556,7 +3556,8 @@ static void mlx5v_shutdown(struct
> > > > > > > > > auxiliary_device
> > > > > > > > > *auxdev)
> > > > > > > > >         mgtdev = auxiliary_get_drvdata(auxdev);
> > > > > > > > >         ndev = mgtdev->ndev;
> > > > > > > > >
> > > > > > > > > -       free_irqs(ndev);
> > > > > > > > > +       if (ndev)
> > > > > > > > > +               free_irqs(ndev);
> > > > > > > > > }
> > > > > > > > >
> > > > > > > >
> > > > > > > > something I don't get:
> > > > > > > > irqs are allocated in mlx5_vdpa_dev_add
> > > > > > > > why are they not freed in mlx5_vdpa_dev_del?
> > > > > > > >
> > > > > > > That is a good point. I will try to find out. I also don't get why
> > > > > > > free_irq
> > > > > > > is
> > > > > > > called in the vdpa dev .free op instead of mlx5_vdpa_dev_del.
> > > > > > > Maybe I
> > > > > > > can
> > > > > > > change
> > > > > > > that in a different refactoring.
> > > > > >
> > > > > > as it is I have no idea whether e.g. ndev can change
> > > > > > between these two call sites. that would make the check
> > > > > > pointless.
> > > > > >
> > > > > > > > this is what's creating all this mess.
> > > > > > > >
> > > > > > > >
> > > > > > > Not quite: mlx5_vdpa_dev_del (which is a .dev_del of for struct
> > > > > > > vdpa_mgmtdev_ops) doesn't get called on shutdown. At least that's
> > > > > > > what
> > > > > > > I
> > > > > > > see. Or
> > > > > > > am I missing something?
> > > > > >
> > > > > > and why do we care whether irqs are freed on shutdown?
> > > > > >
> > > > > Had to ask around a bit to find out the answer: there can be issues
> > > > > with
> > > > > kexec
> > > > > IRQ allocation on some platforms. It is documented here [0] for
> > > > > mlx5_core.
> > > > >
> > > > > https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/net/ethernet/mellanox/mlx5/core/main.c#n2129
> > > > >
> > > > > Thanks,
> > > > > Dragos
> > > >
> > > > It's quite weird.
> > > >          * Some platforms requiring freeing the IRQ's in the shutdown
> > > >          * flow. If they aren't freed they can't be allocated after
> > > >          * kexec. There is no need to cleanup the mlx5_core software
> > > >          * contexts.
> > > >
> > > > but most drivers don't have a shutdown callback how do they work then?
> > > > do you know which platforms these are?
> > >
> > I don't. x86_64 is not one of them though. I will do some more digging ...
> >
Turns out that this fix (releasing the irqs on .shutdown on mlx5_core) was
required for PPC arch but only for certain mainframe systems. That's all the
info I could find.

> > > There used to be bzs that requires virtio drivers to add a shutdown to
> > > fix kexec:
> > >
> > > https://bugzilla.redhat.com/show_bug.cgi?id=2108406
> > >
> > I don't have access to this. What is it about?
>
> This bug might be more accurate:
>
> https://bugzilla.redhat.com/show_bug.cgi?id=1820521
>
> It's about the kexec guys (cced relevant people) wanting to add a
> shutdown method for virito to fix potential kexec issues.
>
> Thanks
>
> >
> > Thanks,
> > Dragos
> > > Thanks
> > >
> > > >
> > > > I don't really know much about why shutdown callback is even necessary.
> > > > I guess this is to detect shutdown and do a faster cleanup than
> > > > the slow, graceful removal, just cleaning hardware resources?
> > > >
.shutdown could be removed in mlx5_vdpa. But I notice that mlx5_core's .shutdown
kicks in from pci_device_shutdown to clean the irqs. So the irqs will still be
freed but as a side effect. Which is not good.

Thanks,
Dragos

Next message: David Rheinsberg: "Re: [PATCH] memfd: support MFD_NOEXEC alongside MFD_EXEC"
Previous message: Daniel Baluta: "Re: [PATCH 1/2] ASoC: simple-card: Introduce playback-only/capture only DAI link flags"
In reply to: Jason Wang: "Re: [PATCH] vdpa/mlx5: Fix crash on shutdown for when no ndev exists"
Next in thread: Dragos Tatulea: "Re: [PATCH] vdpa/mlx5: Fix crash on shutdown for when no ndev exists"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]